The effects of 48 weeks of rosiglitazone on hepatocyte mitochondria in human nonalcoholic steatohepatitis


  • Potential conflict of interest: Nothing to report.


Rosiglitazone, a thiazolidinedione peroxisome proliferator-activated receptor gamma ligand, reduces disease activity in nonalcoholic steatohepatitis (NASH), a disease associated with hepatocyte mitochondrial crystalline inclusions that are not seen in animal models of NASH. In human and animal studies of adipose tissue, thiazolidinediones may induce mitochondrial biogenesis and associated morphological changes. To determine if rosiglitazone alters the hepatocyte mitochondrial morphology in human NASH, we prospectively and systematically examined liver biopsies from human subjects with NASH before and after 48 weeks of rosiglitazone by transmission electron microscopy. Twenty patients (body mass index = 34 ± 7) were studied. Four coded sections from each of 20 pretherapy biopsies and each of 20 posttherapy biopsies were examined by transmission electron microscopy. The total hepatocyte mitochondria and crystal-containing mitochondria were counted, and semiquantitative scoring was performed for macrosteatosis, microsteatosis, dilated endoplasmic reticulum, apoptosis, Mallory bodies, and hepatocyte enlargement. The total mitochondria count was unchanged after therapy, but there was a significant increase in crystal-containing mitochondria from 4.0% (95% confidence interval = 1.8-8.8) to 7.2% (95% confidence interval = 3.9-12.6; odds ratio = 1.80; P = 0.04) after the treatment with rosiglitazone. Macrosteatosis (P < 0.001) and Mallory bodies (P = 0.05) significantly decreased, but no change was evident in microsteatosis, cellular enlargement, dilated endoplasmic reticulum, or apoptosis. Conclusion: Rosiglitazone therapy of NASH is associated with increased crystalline inclusions in hepatocyte mitochondria. Whether these are adaptive or pathological remains unknown, and further studies are warranted to assess hepatic mitochondrial function during thiazolidinedione therapy for NASH. (HEPATOLOGY 2007.)

The presence of impaired mitochondrial oxidative phosphorylation has been known for many years in experimental models of fatty liver disease.1 More recently, hepatocyte mitochondrial dysfunction has also been implicated in the pathogenesis of human nonalcoholic steatohepatitis (NASH).2 In vivo, provocative magnetic resonance spectroscopy testing following fructose-induced adenosine triphosphate (ATP) depletion has been associated with decreased ATP regeneration in human fatty liver disease.3 These changes appear to be related to disturbed function of the mitochondrial electron transport chain and possibly to increased expression of uncoupling protein (UCP) 2.4, 5 Morphologically, the hepatocyte mitochondria in NASH often contain crystalline inclusions (Fig. 1) that are similar ultrastructurally to those described for Wilson's disease and alcohol-related liver disease.6–9 In NASH, these structures are found equally within different zones of the hepatic acini, and their presence correlates to measures of oxidative stress; this suggests a role in disease pathogenesis, although it is not clear if their presence is beneficial or harmful.10, 11 To the best of our knowledge, these structures have not been observed in animal models of NASH.

Figure 1.

Linear mitochondrial crystal inclusions are shown in nearly all mitochondria present (black arrows) along with a fat droplet and foci of the dilated endoplasmic reticulum (asterisk; ×20,000). The inset shows a detail of the crystals that have displaced the normal cristae (×60,000).

Thiazolidinediones (TZDs) are a class of insulin-sensitizing antidiabetic agents that are ligands for the peroxisome proliferator-activated receptor gamma (PPARγ) receptor, a nuclear receptor that is abundantly present in adipose tissue but expressed to a much lesser extent in the normal liver, kidney, and small intestine.12–14 Studies of TZD therapy in human NASH have shown decreased inflammation, cellular ballooning, and hepatic fat observed by light microscopy and improved serum aminotransferase levels.15–17 The reduction in hepatic steatosis has been observed in as little as 6 weeks of therapy.18 However, these beneficial effects are often associated with weight gain, which results from increased peripheral adipose tissue.19, 20 Experimentally, the redistribution of body fat with TZD therapy results from increased fat uptake in peripheral adipose tissue and increased fatty acid oxidation and thermogenesis in visceral fat.21

Because of the mitochondrial abnormalities associated with NASH and the role that mitochondria play in fat metabolism, we hypothesized that the improved light microscopy findings with TZD therapy would be associated with normalization of the mitochondrial morphology. To address this question, we systematically studied the hepatocyte mitochondrial morphology in patients with NASH before and after 48 weeks of rosiglitazone (RSG) therapy. Rather than decreased crystalline inclusions, we observed a statistically significant increase in these structures.


ALT, alanine aminotransferase; ATP, adenosine triphosphate; BMI, body mass index; dER, dilated endoplasmic reticulum; FD, fat droplet; GGT, gamma glutamyl transpeptidase; HgbA1C, glycosylated hemoglobin; MB, Mallory body; NASH, nonalcoholic steatohepatitis; PPARγ, peroxisome proliferator-activated receptor gamma; QUICKI, quantitative insulin sensitivity check index; RSG, rosiglitazone; TEM, transmission electron microscopy; TZD, thiazolidinedione; UCP, uncoupling protein.

Patients and Methods


The study subjects consisted of members of a previously reported cohort of patients with NASH treated with RSG, 4 mg twice daily, for 48 weeks.22 Prospectively collected paired specimens obtained before and after therapy were available for ultrastructural examination in 20 of the 25 patients who completed the treatment (Table 1). The group consisted of 7 adult men and 13 adult women (initial body mass index = 34 ± 7.4 kg/m2), of whom 8 had diabetes and 6 had impaired glucose tolerance. As previously reported, a blinded light microscopy evaluation demonstrated that RSG therapy was associated with significantly reduced inflammation, ballooning, perisinusoidal fibrosis, and hepatic steatosis, the last measured both histologically and globally by computerized tomographic (CT) measured liver/spleen density ratio.

Table 1. Laboratory Parameters (Means and Ranges) Before and After 48 Weeks of Rosiglitazone Therapy
  • ALT indicates alanine aminotransferase; BMI, body mass index; GGT, gamma glutamyl transpeptidase; HgbA1C, glycosylated hemoglobin; and QUICKI, quantitative insulin sensitivity check index.

  • *

    Significant differences after therapy (P < 0.05).

BMI (kg/m2)33.8 (26.3–60.2)36.3 (27.7–64.9)*
ALT (U/L)96.6 (29–332)41.7 (14–129)*
GGT (U/L)101 (41–275)35 (19–88)*
Cholesterol (mg/dL)217 (152–294)222 (147–314)
Glucose (mg/dL)124 (77–276)103 (84–133)
HgbA1C (%)5.92 (4.8–8.1)5.43 (4.6–6.6)*
QUICKI0.287 (0.255–0.321)0.318 (0.268–0.354)*

Specimen Preparation.

At the initial and posttreatment biopsies, 2-mm segments of liver tissue were separated from the core and immediately fixed in a solution containing 4.0% (wt/vol) paraformaldehyde and 2.0% (wt/vol) glutaraldehyde in a 0.1 M phosphate buffer (pH 7.2, 24°C). The samples were fixed for 1 hour in 1.0% osmium tetroxide at 24°C, dehydrated in ethanol, and embedded in a resin for transmission electron microscopy (TEM); this resulted in 2 blocks of tissue for each biopsy. The specimen blocks, obtained after fixation and embedding at Saint Louis University, were coded for a blinded evaluation and submitted for sectioning and assessment at the University of Virginia in accordance with investigational review board protocols approved at both universities. Two 0.5-μm sections (thick sections), separated by 30 μm, were taken from each block, providing 2 nonoverlapping levels per block. The thick sections were stained with toluidine blue for examination by light microscopy to confirm the adequacy of the samples and to identify the primary regions of the hepatic lobule from which the specimens were derived, as detailed later. Two ultrathin sections (70-80 nm) were immediately cut serially to the 0.5-μm-thick sections, and each was placed on a 200-mesh copper grid, contrast-stained with lead citrate and uranyl acetate, and examined with a JEOL 100-CX transmission electron microscope. From each grid (1 grid per section, 2 grids per block, and 4 grids per biopsy), 10 grid holes were examined (in a blinded fashion to pretherapy or posttherapy), and this resulted in the examination of 40 grid holes per patient before therapy and 40 grid holes per patient after therapy. Each grid was examined similarly: the examination was begun in the left upper corner, and only grid holes that were completely covered with tissue were counted. This resulted in the systematic examination of 797 grid holes before RSG therapy and 800 grid holes after RSG therapy. These methods are illustrated in Fig. 2.

Figure 2.

Study design: the pretreatment and posttreatment biopsies were fixed, embedded, and sectioned to provide 4 different grids per biopsy, which were used for quantitative and semiquantitative assessments.


A quantitative analysis was performed as follows: the total number of mitochondria and the total number of crystal-containing mitochondria were counted from ×10,000 TEM negatives taken at the center of each of the counted grid holes. The area of the negative at this magnification was approximately 70 μm2. A semiquantitative assessment was performed in each grid hole for the following ultrastructural features and was scored as 0 if the feature was absent or 1 if it was present: macrosteatosis (defined as a single dominant lipid droplet within a hepatocyte with peripheral displacement of the nucleus), microsteatosis (defined as >4 fat droplets within the cell without peripheral displacement of the nucleus), dilated endoplasmic reticulum (dER), apoptosis (nuclear condensation or blebbing of the cell membrane), and Mallory body formation. We also sought to assess hepatocyte ballooning. Because there is no consensus definition of the ultrastructural features of hepatocyte ballooning in NASH, we chose to use a size criterion (hepatocyte diameter > 30 μm) to assess this aspect of NASH, recognizing that there is normal variation in the cell size within the lobule and that a more precise definition of cellular ballooning is an area of controversy.23–25 In secondary analyses, light microscopy of the toluidine blue–stained thick sections was used to determine the primary origin of the resin-embedded specimen from within the hepatic lobule as follows: positive identification of structures of the portal tract, zone 1, and identification of an isolated central vein, zone 3. If neither of these structures could be clearly identified, the specimen was identified as indeterminate.


The pretherapy and posttherapy data for the total mitochondria, crystal-containing mitochondria, and other semiquantitative measurements were summarized as counts and frequencies. Multivariate negative-binomial and multivariate binomial generalized estimating equation models were constructed to analyze these data. A regional analysis was performed with pooled data from each of the 3 regions before and after therapy. The pretherapy and posttherapy counts and frequencies were compared by way of linear contrasts. The working variance-covariance parameter estimates from the generalized estimating equation model were used to derive the Wald statistic for the hypothesis test. Confidence interval construction was likewise based on the Wald statistic. We used the PROC GENMOD procedure of SAS, version 9.1 (SAS Institute Inc., Cary, NC), to conduct these analyses.


The clinical parameters of the cohort described here before and after 48 weeks of RSG are shown in Table 1. As previously reported, RSG therapy was associated with significant improvements in steatosis, necroinflammation, and fibrosis in a blinded review of the formalin fixed specimens.22 A comparison of the pretreatment and posttreatment paired samples for changes by TEM showed no difference in the total number of mitochondria per unit of area after the treatment: 24.64 (95% confidence interval = 23.10-26.28) versus 26.03 (95% confidence interval = 24.36-27.83). However, there was a significant increase in the percentage of mitochondria containing crystal inclusions after therapy, from 4.0% (95% confidence interval = 1.8-8.8) of all mitochondria before the treatment to 7.2% (95% confidence interval = 3.9-12.6; odds ratio = 1.80; P = 0.04) after RSG (Fig. 3). On the basis of a semiquantitative analysis, a comparison of paired samples revealed a significant reduction in the odds of macrosteatosis (P < 0.001) and Mallory bodies (P = 0.047) after RSG (Fig. 4). We could detect no statistically significant changes in the frequency of microsteatotic hepatocytes, hepatocytes with dER, or hepatocytes greater than 30 μ in diameter. Ultrastructural evidence of apoptosis (membrane blebbing and nuclear condensation) was absent in all samples (before and after therapy), indicating either rapid clearance and/or the presence of a balance between increased pro-apoptotic and anti-apoptotic factors.26, 27

Figure 3.

The percentage of mitochondria containing crystal inclusions before therapy was 4.0% (95% confidence interval = 1.8-8.8) versus 7.2% (95% confidence interval = 3.9-12.6) after therapy (odds ratio = 1.80; P = 0.04). The confidence intervals are shown and were estimated by the transformation of the lower and upper limits of the 95% confidence interval for the log odds to the probability scale. With this technique, asymmetry of the confidence intervals is expected.

Figure 4.

(A) The probability of observing macrosteatosis after therapy was significantly reduced (P = 0.001). (B) The probability of observing Mallory bodies was reduced after rosiglitazone therapy (P = 0.047). The vertical bars represent 95% confidence intervals for the probability of the event. The confidence intervals were estimated by the transformation of the lower and upper limits of the 95% confidence interval for the log odds to the probability scale. With this technique, asymmetry of the confidence intervals is expected.

When the sections were classified by the lobular region, there was a comparable distribution of the 40 pretreatment sections and the 40 posttreatment sections. In the pretreatment specimens, 9 sections were identified as zone 1, 9 were identified as zone 3, and 22 were indeterminate. After therapy, 10 were identified as zone 1, 7 were identified as zone 3, and 23 were indeterminate. Pooled, unpaired semiquantitative scoring by the dominant lobular region (zone 1, zone 3, or indeterminate) showed no regional variation in the distribution of mitochondria with crystalline inclusions, and this is consistent with previously reported observations.11 In the pretreatment samples, hepatocyte enlargement and dER were both significantly more frequent in non–zone 1 regions, but these zonal differences were not evident after therapy. Macrosteatosis declined significantly with therapy in all 3 zones (P < 0.01 in zones 1 and 3 and P = 0.05 in the indeterminate group, Wald test). dER was commonly seen before and after RSG therapy and was often associated with multiple small fat droplets in enlarged cells (Fig. 5).

Figure 5.

(A) Enlarged hepatocyte with multiple small fat droplets and a nonperipherally displaced nucleus consistent with microsteatosis. (B) Convergence (arrow) of the dilated endoplasmic reticulum and a fat droplet. (C) Numerous tangled filaments of a Mallory body in proximity to a fat droplet (magnification, ×60,000). dER indicates dilated endoplasmic reticulum; FD, fat droplet; and MB, Mallory body.


Although we anticipated normalization of the mitochondrial morphology in concert with improved light microscopy histology, we observed a significant increase in hepatocyte mitochondria containing crystalline inclusions after RSG therapy. This finding is consistent with a previous study of troglitazone, an older TZD that was associated with rare cases of hepatic necrosis in humans and also increased the number of mitochondria containing crystalline inclusions while ameliorating light microscopy injury.28 Although the mechanism remains to be defined, the findings indicate that these structures may be part of a beneficial adaptive response that is further augmented by TZD therapy in human fatty liver disease.

Intramitochondrial crystals occur in human NASH as bundles of parallel strands (Fig. 1) seen in 5%-15% of hepatocytes.7, 28 Previous optical diffraction studies have suggested that these structures are composed of a lipid material.6 Although their composition remains unresolved, there is increasing evidence supporting a link between these structures and disturbed mitochondrial energy homeostasis. A number of studies have demonstrated impaired mitochondrial respiratory chain function in NASH in both animal and human studies.2, 4, 29 Moreover, the inhibition of complex 1 of the electron transport chain has been reported with TZD exposure, and the inhibition of this subunit in patients treated with nucleoside reverse transcriptase inhibitors has been correlated with the presence of similar hepatocyte mitochondrial inclusions.30, 31

Although there is species variation, PPARγ agonists usually cause a shift from central fat depots to peripheral fat depots.32 For example, human studies of PPARγ agonists in NASH have consistently shown decreased hepatic fat and concomitant peripheral weight gain.15 In Sprague-Dawley rats fed a high-sucrose, high-fat diet, this results from a greater increase in energy expenditure in visceral fat.33 These changes are associated with TZD-stimulated mitochondrial biogenesis, which has been demonstrated in vitro in human white adipocyte cell cultures and in vivo in samples of human subcutaneous adipose tissue obtained after 12 weeks of pioglitazone.34, 35 Stimulation of mitochondrial proliferation may be mediated by PPARγ receptor activation directly or independently through the activation of adenosine monophosphate–activated protein kinase, which has been shown in rat adipose tissue, livers, and skeletal muscle.36, 37 The latter pathway may be especially relevant because of the association of TZD therapy with increased adiponectin, which activates adenosine monophosphate–activated protein kinase in humans.38 Increased mitochondria with crystal inclusions in the human liver may represent another form of TZD-stimulated mitochondrial biogenesis. Alternatively, pioglitazone and RSG increase UCP in human adipocytes, and RSG significantly alters the mitochondrial morphology in adipocyte cell lines with increased brown fat morphology.34, 39 Both UCP expression and PPARγ receptors are increased in the liver in NASH.40, 41 TZD therapy may augment this process in human NASH with secondary changes in the mitochondrial morphology.

Our study has several limitations. Liver biopsy specimens represent a small portion of the liver, and TEM focuses on an even smaller portion of the biopsy. Thus, sampling error could have been a factor in our results. However, when it was possible to compare measured parameters by different techniques, we noted consistency in the results. For example, these patients showed reduced steatosis by light microscopy, by computed tomography imaging, and by TEM. Another potential problem is the uncertainty of whether mitochondria visualized by TEM represent branches of a few interconnected mitochondria cut in different planes or multiple, discrete organelles. On the basis of previous experience, in which the majority but not all crystal-containing mitochondria appeared swollen, we chose to assess mitochondria with crystals, regardless of the presence or absence of apparent organelle swelling. In addition, because of limitations in obtaining human liver tissue, we did not obtain additional tissue samples to perform biochemical analyses. However, related studies of energy homeostasis using P31 magnetic resonance spectroscopy in vivo to assess ATP metabolism are underway in both TZD and non–TZD-treated NASH patients, as is another study examining PPARγ expression in the human fatty liver and HCC, and tissue is being collected currently for a biochemical study of fat metabolic pathways in treated NASH patients. The results of this article support the need for a close examination of hepatic mitochondrial metabolism in human fatty liver disease.

In summary, we observed no change in the absolute number of mitochondria, but the number of mitochondria that contained crystalline inclusions significantly increased in association with decreased macrosteatosis and decreased Mallory body frequency after the RSG treatment. TZD therapy was also associated with changes in the lobular distribution of cells with dER and the distribution of enlarged hepatocytes after therapy. On the basis of the effects of TZDs on adipocyte metabolism, changes in the hepatocyte mitochondrial morphology most likely represent altered hepatocyte energy homeostasis. Further studies are warranted to better understand these relationships.